LOOP FORGE // PROMPT TERMINAL
BUILD. CRITIQUE. UPGRADE. REPEAT.

Raise the bar.
Build the loop.

Give your agent a goal, a real benchmark, and an independent critic. Keep the work moving until it earns a pass.

01 Set the direction

You choose the goal.
A destination. Let the agent choose the route.0 / 12,000
TRY

Use something the critic can inspect. Leave blank to get a suggested bar.

Constraints & stopping budget

Adds a stop instruction to the prompt. This app does not run agents or enforce spend.

Instant template generation Ctrl / ⌘ + Enter

02 Give it a higher bar

YOUR LOOP
QUALITY BAR SUGGESTED · NOT YET VERIFIED

Start with your brief. We’ll suggest a concrete way to judge the result.

01Build
02Critique
03Improve

Ambition, with a feedback loop.

Your ready-to-run prompt will appear here.
Each builder gets a fresh pair of critical eyes.

CONCRETE BAR · REAL EVIDENCE · INDEPENDENT CRITIC
THE METHOD

A loop that earns its pass.

Start with a result you want and evidence that would prove it is good. Let a lead agent choose how to get there.

  1. Find an inspectable bar.A rendered reference, a working product, a strong writing sample, or a measurable test. A suggestion becomes a bar only after the agent obtains the evidence.
  2. Split work where it makes sense.Let the lead choose independently judgeable pieces. Coordinate changes that touch the same systems.
  3. Give the critic fresh context.Share the goal, bar, constraints, and actual artifact. Keep the builder’s explanation out of the judgment.
  4. Close the biggest gap. Repeat.Inspect, compare, revise. Keep an honest progress record. Stop when the bar is met, you stop the run, or your supplied budget is reached.

Loop Forge uses local templates and task detection. It has no email gate, remote generation queue, or model calls. History is stored in this browser. It cannot guarantee that an agent follows the prompt or reaches the benchmark.

Published method ↗Original game prompt ↗
IN THIS BROWSER

Saved loops